A queryable in-memory index is created over the files in the data directory upon startup.
New documents can be added to the index during runtime by sending a POST to the /document endpoint.
The default index used is the in-memory GPTSimpleVectorIndex. The vector db Pinecone is also supported.
To use Pinecone, set the INDEX_TYPE env var to pinecone and populate the PINECONE_API_KEY env var.
The default LLM used is flan-t5-large, an Apache-2.0-licensed LLM that is downloaded once upon startup and then runs
locally.
To instead use gpt-3.5-turbo, set the LLM_MODEL_NAME env var to gpt-3.5-turbo.
If gpt-3.5-turbo is used, the OPENAI_API_KEY env var must also be set.
Install dependencies with pip install -r requirements.txt
Run with python -m uvicorn app:app --reload
main.http contains sample HTTP requests for the REST API.